Aptos processed 111.9 user transactions a second over a measured 25 minutes this morning. Ninety-four per cent of them went to one application, a perpetual futures exchange called Decibel, and 85 per cent came from three accounts. The chain is fast and it works. What it mostly carries is one exchange’s bots talking to its own order book.
The method: we read the ledger version and block height from two independent public nodes, api.mainnet.aptoslabs.com and aptos-rest.publicnode.com, at 06:07:44 and 06:33:00 UTC, then sampled 8,000 transactions spread evenly across the window. Both nodes gave 164.98 ledger versions a second, and eight spot-checked transaction hashes matched.
What the numbers are counting
Ledger versions are not user transactions. Every Aptos block carries two system entries, a block-metadata record and a block epilogue, and at 26.54 blocks a second that is 53.1 entries a second of pure bookkeeping. Take those out and 111.9 user transactions a second remain; the independent sample gives 110.6.
Decode those and the picture narrows. The largest single sender, 45.6 per cent of user traffic, is a bot placing and repricing bulk orders on Decibel. The second, 31.3 per cent, does nothing but post Chainlink-derived mark prices to the same exchange. The third, 8.2 per cent, is a keeper processing its pending requests. Everything on Aptos that is not Decibel comes to about seven transactions a second.
The claims, and the docs
The network’s own site, aptosnetwork.com, promises “Sub-second finality” and “Zero downtime”. The first stands up: mean block time was 37.7 milliseconds, the median 25.5, and the longest gap in 40,222 consecutive blocks was 602 milliseconds. The second does not. Aptos Labs’ own archive node records block 104,621,347 at 23:11:58 UTC on 18 October 2023 and the next at 04:17:54 on 19 October: five hours, five minutes and 55 seconds without a block.
The much-quoted 160,000 figure reaches the developer docs only as a link to a Medium post whose address reads “block-stm-how-we-execute-over-160k-transactions-per-second”. That is an execution-engine benchmark, not network throughput, and the same docs ask node operators for hardware able to “maintain ~30,000 transactions per second”. Measured user traffic is 0.07 per cent of the first number and 0.37 per cent of the second.
| Ledger versions per second | 164.98 |
| User transactions per second | 111.9 |
| Share sent to Decibel | 93.7% |
| Non-Decibel user traffic | about 7 per second |
| Mean block time; longest gap | 37.7 ms; 602 ms |
| October 2023 halt, from block timestamps | 5 h 5 min 55 s |
| DeFi TVL; Decibel’s share | $56.8m; 52.1% |
| Stablecoins on chain | $1.12bn |
| APT price; fully diluted value | £0.552; £668m (CoinGecko) |
| Validators to exceed one third of stake | 14 of 85 |
Aptos can clearly do more than 112 transactions a second. Almost nobody outside one exchange is asking it to.
What has shipped
Read the consensus configuration stored on chain and it names JolteonV2, with quorum store switched on. Raptr, the consensus redesign behind the lower-latency roadmap, is not an option in the released code; a blog post announces a “Baby Raptr” step we could not fetch and so cannot confirm. The encrypted-transactions feature flag is off, although the docs already describe an encrypted mempool. Keyless accounts and on-chain randomness are on, and randomness key-generation transactions appeared at this morning’s epoch change.
The bear case
DeFi on Aptos is small and concentrated. DefiLlama puts value locked at $56.8m, over half of it in Decibel, and ThalaSwap V3 handled 74 per cent of the day’s $5.68m of exchange volume. The chain’s own gas revenue was $4,164 in 24 hours, which our sampled fees independently put at about $4,500. A £481m market capitalisation rests on activity that one venue could take elsewhere.
Supply is the other weight. CoinGecko lists 870.6m APT circulating of 1,209m in existence, and quotes a 2.1bn maximum that the on-chain supply limit does not enforce.
The bull case
The engineering is real. Blocks every few tens of milliseconds, sustained, on 85 validators with the largest holding 3.4 per cent of stake, is a better result than most chains making larger claims. It takes 14 validators to pass a third of voting power and 36 to pass two thirds, and 63 per cent of all APT is staked.
And $1.12bn of stablecoins sits on the chain, about twenty times its DeFi value locked. If the money that is parked ever starts to move, the capacity to carry it is demonstrably there.
Sources, all retrieved 21 September 2026 between 06:07 and 06:50 UTC: throughput, block times, transaction types, senders and payload functions measured directly against api.mainnet.aptoslabs.com and aptos-rest.publicnode.com over a 1,515-second window on the chain clock, with the Aptos indexer for block timestamps; the October 2023 halt from archive.mainnet.aptoslabs.com block timestamps; consensus configuration, feature flags, validator set and supply from on-chain resources at 0x1; aptosnetwork.com and aptos.dev for the claims, quoted as claims; DefiLlama for TVL, stablecoins, DEX volume and fees; CoinGecko for price and supply, using its own sterling quote. TVL and fees are in dollars. Contested: “TPS” is 165 or 112 depending on whether system entries are counted, and both are shown; TVL is $56.8m on DefiLlama’s chain figure, $77.0m summing protocols and $425m with exchange wallets; DefiLlama’s per-protocol fee rows sum to more than its own total; validator counts are per validator, and grouping by operator could lower them. The aptosfoundation.org homepage made no numerical claim when fetched. A 25-minute window is a snapshot, not a daily average. Not a recommendation to buy, sell or hold anything.